Discovering Speed Changes of Vehicles from Audio Data

Area of Science:

• Acoustic analysis
• Machine learning for transportation safety

Background:

• Monitoring driver behavior near control points is crucial for traffic safety.
• Objective assessment of vehicle speed changes requires reliable detection methods.

Purpose of the Study:

• To develop and evaluate a system for detecting speed changes (acceleration, deceleration, constant speed) from car audio recordings.
• To assess driver behavior near traffic control points for safety compliance.

Main Methods:

• Utilized Support Vector Machines (SVM), random forests, and artificial neural networks for classification.
• Explored various audio data representations including feature averages, time evolution, and spectrogram analysis.
• Developed an ensemble classifier combining the best performing models.

Main Results:

• Achieved nearly 95% classification accuracy for speed changes.
• Demonstrated high precision in distinguishing acceleration from deceleration.
• Showcased minimal errors between stable speed and speed variations.

Conclusions:

• The proposed audio-based system effectively detects vehicle speed changes.
• This technology offers a foundation for developing traffic safety campaigns and monitoring systems.